Training on the custom dataset?

Hello, I really liked the idea of your paper. I am interested in using this model for another dataset with NER. Is it possible to use my own dataset in the same format with a bigger sequence length (500-2000 tokens)? Or is this model and implementation is not suitable for this purpose?